Add XNOR Net about zoo HOT 6 CLOSED

Note that in the original implementation, weight decay is used instead that we know is similar, but different to L2.

If you want weight decay with Adam or SGD you can copy the implementation from here: https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/weight_decay_optimizers.py#L341

from zoo.

See branch on Larq Models, WIP

from zoo.

This has been ported over to it's own repository and is WIP. Currently validation accuracy sits around 42.5% and model is sensitive to the value of L2 normalisation chosen. Note that in the original implementation, weight decay is used instead that we know is similar, but different to L2.

from zoo.

Currently have got to 43.3% Validation Accuracy (Paper is 44.2%). As mentioned before, model is very sensitive to the value for L2 regularization. Training accuracy has gotten as high as 47% now so there looks like there is definitely wiggle room to get to the target accuracy

from zoo.

See #26 for draft PR

from zoo.

Closed in #26

from zoo.